This paper presents the work we have conducted concerning real time scheduling in Camille, an exokernel dedicated to smart cards. We show that it is possible to embedded a flexible real-time operating system despite the important hardware limitations of the smart card platform. We present the major difficulties one has to face when integrating real time support in an exokernel embedded on a very resource-limited platform. We first present a naive solution consisting in allocating an equal time slice to every system extensions and letting each one share it as needed amongst its tasks. We show that this solution does not account for loading of new extensions in the system, and that it can fail if some extensions have much more work to carry out than the others. We then present a more complex solution based upon collaborative schedulers grouped as virtual extensions. We show that this solution supports dynamic loading of new extensions and works even for very unbalanced task repartitions. We finally address the issue of trust between the collaborating extensions and we propose a solution based on exhaustive testing and formal proving of the plan functions.